Dashboard › opencode-lore › Distillation
1e946659-8c31-48de-bcf9-31affbabcc41["lore_tm_v1_QwMLYpnlod9TCS6kLX1SdLDCojjEFpsk1Sb_1bMd1JM","lore_tm_v1_Fw06NfMiQS3HL8BIYJTVufb_hPLdkC5JUZUAm1cHrmA","lore_tm_v1_Gng1_LJznTxb5wbDMgxuwFC56z0oPyTw2vnMWMYVQ3Q","lore_tm_v1_xy4WJAvo4BCCX2ozHCZxhYGTw3Uo0qRpuvirQC1IZ5M","lore_tm_v1_AEK11jkRUFHGTjdArh8PXpZegAQ946CblclWs6KkoO8"]
Date: Sep 16, 2026
packages/gateway/test/codex-usage-metadata.test.ts defines request-wide Codex quota metadata caps of MAX_CODEX_RATE_LIMIT_EVENTS = 64 and MAX_CODEX_RATE_LIMIT_BYTES = 16 * 1024.packages/gateway/test/codex-usage-metadata.test.ts verifies buffered rebuilding preserves every ordered codex.rate_limits bucket before response.completed, including on ResponsesTerminalError from an incomplete terminal.authorization, debug, codex.private_metadata, unknown nested rate-limit buckets, and private credit fields—from both buffered and passthrough output.packages/gateway/test/codex-usage-metadata.test.ts run seeds 0 through 9, each with numRuns: 25, generating 1–20 ordered quota updates; they verify exact update order and unchanged consumption metadata, including inputTokens = 600000.packages/gateway/test/openai-responses-recall-aware-stream.test.ts identify coverage for: retrying a principal read failure before any client event; continuing after a post-text read failure without replaying the request; transactionally failing final continuations without retry; avoiding double-charging transactional output during continuation retry; restoring every continuation retry budget and identity baseline; and suppressing retry when cancellation races a dropped continuation.maxDeferredBytes configurations including 2 * 1024 * 1024 and 16 * 1024; failures are collected through setRecallContinuationFailureHook() using RecallContinuationFailureCategory.